8 种经常被忽视的 SQL 错误用法,你有没有踩过坑…

2019-11-18 09:22:30来源:博客园 阅读 ()

新老客户大回馈,云服务器低至5折

8 种经常被忽视的 SQL 错误用法,你有没有踩过坑?

1、LIMIT 语句

分页查询是最常用的场景之一,但也通常也是最容易出问题的地方。比如对于下面简单的语句,一般 DBA 想到的办法是在 type, name, create_time 字段上加组合索引。这样条件排序都能有效的利用到索引,性能迅速提升。

SELECT * 
FROM   operation 
WHERE  type = 'SQLStats' 
       AND name = 'SlowLog' 
ORDER  BY create_time 
LIMIT  1000, 10;

好吧,可能 90% 以上的 DBA 解决该问题就到此为止。但当 LIMIT 子句变成 “LIMIT 1000000,10” 时,程序员仍然会抱怨:我只取 10 条记录为什么还是慢?

要知道数据库也并不知道第 1000000 条记录从什么地方开始,即使有索引也需要从头计算一次。出现这种性能问题,多数情形下是程序员偷懒了。

在前端数据浏览翻页,或者大数据分批导出等场景下,是可以将上一页的最大值当成参数作为查询条件的。SQL 重新设计如下:

SELECT   * 
FROM     operation 
WHERE    type = 'SQLStats' 
AND      name = 'SlowLog' 
AND      create_time > '2017-03-16 14:00:00' 
ORDER BY create_time limit 10;

在新设计下查询时间基本固定,不会随着数据量的增长而发生变化。

2、隐式转换

SQL 语句中查询变量和字段定义类型不匹配是另一个常见的错误。比如下面的语句:

mysql> explain extended SELECT * 
     > FROM   my_balance b 
     > WHERE  b.bpn = 14000000123 
     >       AND b.isverified IS NULL ;
mysql> show warnings;
| Warning | 1739 | Cannot use ref access on index 'bpn' due to type or collation conversion on field 'bpn'

其中字段 bpn 的定义为 varchar(20),MySQL 的策略是将字符串转换为数字之后再比较。函数作用于表字段,索引失效。

上述情况可能是应用程序框架自动填入的参数,而不是程序员的原意。现在应用框架很多很繁杂,使用方便的同时也小心它可能给自己挖坑。

3、关联更新、删除

虽然 MySQL5.6 引入了物化特性,但需要特别注意它目前仅仅针对查询语句的优化。对于更新或删除需要手工重写成 JOIN。

比如下面 UPDATE 语句,MySQL 实际执行的是循环 / 嵌套子查询(DEPENDENT SUBQUERY),其执行时间可想而知。

UPDATE operation o 
SET    status = 'applying' 
WHERE  o.id IN (SELECT id 
                FROM   (SELECT o.id, 
                               o.status 
                        FROM   operation o 
                        WHERE  o.group = 123 
                               AND o.status NOT IN ( 'done' ) 
                        ORDER  BY o.parent, 
                                  o.id 
                        LIMIT  1) t);


执行计划:

+----+--------------------+-------+-------+---------------+---------+---------+-------+------+-----------------------------------------------------+
| id | select_type        | table | type  | possible_keys | key     | key_len | ref   | rows | Extra                                               |
+----+--------------------+-------+-------+---------------+---------+---------+-------+------+-----------------------------------------------------+
| 1  | PRIMARY            | o     | index |               | PRIMARY | 8       |       | 24   | Using where; Using temporary                        |
| 2  | DEPENDENT SUBQUERY |       |       |               |         |         |       |      | Impossible WHERE noticed after reading const tables |
| 3  | DERIVED            | o     | ref   | idx_2,idx_5   | idx_5   | 8       | const | 1    | Using where; Using filesort                         |
+----+--------------------+-------+-------+---------------+---------+---------+-------+------+-----------------------------------------------------+

重写为 JOIN 之后,子查询的选择模式从 DEPENDENT SUBQUERY 变成 DERIVED,执行速度大大加快,从 7 秒降低到 2 毫秒

UPDATE operation o 
       JOIN  (SELECT o.id, 
                            o.status 
                     FROM   operation o 
                     WHERE  o.group = 123 
                            AND o.status NOT IN ( 'done' ) 
                     ORDER  BY o.parent, 
                               o.id 
                     LIMIT  1) t
         ON o.id = t.id 
SET    status = 'applying' 

执行计划简化为:

+----+-------------+-------+------+---------------+-------+---------+-------+------+-----------------------------------------------------+
| id | select_type | table | type | possible_keys | key   | key_len | ref   | rows | Extra                                               |
+----+-------------+-------+------+---------------+-------+---------+-------+------+-----------------------------------------------------+
| 1  | PRIMARY     |       |      |               |       |         |       |      | Impossible WHERE noticed after reading const tables |
| 2  | DERIVED     | o     | ref  | idx_2,idx_5   | idx_5 | 8       | const | 1    | Using where; Using filesort                         |
+----+-------------+-------+------+---------------+-------+---------+-------+------+-----------------------------------------------------+

4、混合排序

MySQL 不能利用索引进行混合排序。但在某些场景,还是有机会使用特殊方法提升性能的。

SELECT * 
FROM   my_order o 
       INNER JOIN my_appraise a ON a.orderid = o.id 
ORDER  BY a.is_reply ASC, 
          a.appraise_time DESC 
LIMIT  0, 20 


执行计划显示为全表扫描:

+----+-------------+-------+--------+-------------+---------+---------+---------------+---------+-+
| id | select_type | table | type   | possible_keys     | key     | key_len | ref      | rows    | Extra    
+----+-------------+-------+--------+-------------+---------+---------+---------------+---------+-+
|  1 | SIMPLE      | a     | ALL    | idx_orderid | NULL    | NULL    | NULL    | 1967647 | Using filesort |
|  1 | SIMPLE      | o     | eq_ref | PRIMARY     | PRIMARY | 122     | a.orderid |       1 | NULL           |
+----+-------------+-------+--------+---------+---------+---------+-----------------+---------+-+

由于 is_reply 只有 0 和 1 两种状态,我们按照下面的方法重写后,执行时间从 1.58 秒降低到 2 毫秒。

SELECT * 
FROM   ((SELECT *
         FROM   my_order o 
                INNER JOIN my_appraise a 
                        ON a.orderid = o.id 
                           AND is_reply = 0 
         ORDER  BY appraise_time DESC 
         LIMIT  0, 20) 
        UNION ALL 
        (SELECT *
         FROM   my_order o 
                INNER JOIN my_appraise a 
                        ON a.orderid = o.id 
                           AND is_reply = 1 
         ORDER  BY appraise_time DESC 
         LIMIT  0, 20)) t 
ORDER  BY  is_reply ASC, 
          appraisetime DESC 
LIMIT  20;

5、EXISTS语句

MySQL 对待 EXISTS 子句时,仍然采用嵌套子查询的执行方式。如下面的 SQL 语句:

SELECT *
FROM   my_neighbor n 
       LEFT JOIN my_neighbor_apply sra 
              ON n.id = sra.neighbor_id 
                 AND sra.user_id = 'xxx' 
WHERE  n.topic_status < 4 
       AND EXISTS(SELECT 1 
                  FROM   message_info m 
                  WHERE  n.id = m.neighbor_id 
                         AND m.inuser = 'xxx') 
       AND n.topic_type <> 5 

执行计划为:

+----+--------------------+-------+------+-----+------------------------------------------+---------+-------+---------+ -----+
| id | select_type        | table | type | possible_keys     | key   | key_len | ref   | rows    | Extra   |
+----+--------------------+-------+------+ -----+------------------------------------------+---------+-------+---------+ -----+
|  1 | PRIMARY            | n     | ALL  |  | NULL     | NULL    | NULL  | 1086041 | Using where                   |
|  1 | PRIMARY            | sra   | ref  |  | idx_user_id | 123     | const |       1 | Using where          |
|  2 | DEPENDENT SUBQUERY | m     | ref  |  | idx_message_info   | 122     | const |       1 | Using index condition; Using where |
+----+--------------------+-------+------+ -----+------------------------------------------+---------+-------+---------+ -----+ 

去掉 exists 更改为 join,能够避免嵌套子查询,将执行时间从 1.93 秒降低为 1 毫秒。

SELECT *
FROM   my_neighbor n 
       INNER JOIN message_info m 
               ON n.id = m.neighbor_id 
                  AND m.inuser = 'xxx' 
       LEFT JOIN my_neighbor_apply sra 
              ON n.id = sra.neighbor_id 
                 AND sra.user_id = 'xxx' 
WHERE  n.topic_status < 4 
       AND n.topic_type <> 5 

新的执行计划:

+----+-------------+-------+--------+ -----+------------------------------------------+---------+ -----+------+ -----+
| id | select_type | table | type   | possible_keys     | key       | key_len | ref   | rows | Extra                 |
+----+-------------+-------+--------+ -----+------------------------------------------+---------+ -----+------+ -----+
|  1 | SIMPLE      | m     | ref    | | idx_message_info   | 122     | const    |    1 | Using index condition |
|  1 | SIMPLE      | n     | eq_ref | | PRIMARY   | 122     | ighbor_id |    1 | Using where      |
|  1 | SIMPLE      | sra   | ref    | | idx_user_id | 123     | const     |    1 | Using where           |
+----+-------------+-------+--------+ -----+------------------------------------------+---------+ -----+------+ -----+

6、条件下推

外部查询条件不能够下推到复杂的视图或子查询的情况有:

  • 聚合子查询;

  • 含有 LIMIT 的子查询;

  • UNION 或 UNION ALL 子查询;

  • 输出字段中的子查询;

如下面的语句,从执行计划可以看出其条件作用于聚合子查询之后

 SELECT * 
FROM   (SELECT target, 
               Count(*) 
        FROM   operation 
        GROUP  BY target) t 
WHERE  target = 'rm-xxxx' 
+----+-------------+------------+-------+---------------+-------------+---------+-------+------+-------------+
| id | select_type | table      | type  | possible_keys | key         | key_len | ref   | rows | Extra       |
+----+-------------+------------+-------+---------------+-------------+---------+-------+------+-------------+
|  1 | PRIMARY     | <derived2> | ref   | <auto_key0>   | <auto_key0> | 514     | const |    2 | Using where |
|  2 | DERIVED     | operation  | index | idx_4         | idx_4       | 519     | NULL  |   20 | Using index |
+----+-------------+------------+-------+---------------+-------------+---------+-------+------+-------------+

确定从语义上查询条件可以直接下推后,重写如下:

SELECT target, 
       Count(*) 
FROM   operation 
WHERE  target = 'rm-xxxx' 
GROUP  BY target

执行计划变为:

+----+-------------+-----------+------+---------------+-------+---------+-------+------+--------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-----------+------+---------------+-------+---------+-------+------+--------------------+
| 1 | SIMPLE | operation | ref | idx_4 | idx_4 | 514 | const | 1 | Using where; Using index |
+----+-------------+-----------+------+---------------+-------+---------+-------+------+--------------------+


7、提前缩小范围

先上初始 SQL 语句:

SELECT * 
FROM   my_order o 
       LEFT JOIN my_userinfo u 
              ON o.uid = u.uid
       LEFT JOIN my_productinfo p 
              ON o.pid = p.pid 
WHERE  ( o.display = 0 ) 
       AND ( o.ostaus = 1 ) 
ORDER  BY o.selltime DESC 
LIMIT  0, 15 

该 SQL 语句原意是:先做一系列的左连接,然后排序取前 15 条记录。从执行计划也可以看出,最后一步估算排序记录数为 90 万,时间消耗为 12 秒。

+----+-------------+-------+--------+---------------+---------+---------+-----------------+--------+----------------------------------------------------+
| id | select_type | table | type   | possible_keys | key     | key_len | ref             | rows   | Extra                                              |
+----+-------------+-------+--------+---------------+---------+---------+-----------------+--------+----------------------------------------------------+
|  1 | SIMPLE      | o     | ALL    | NULL          | NULL    | NULL    | NULL            | 909119 | Using where; Using temporary; Using filesort       |
|  1 | SIMPLE      | u     | eq_ref | PRIMARY       | PRIMARY | 4       | o.uid |      1 | NULL                                               |
|  1 | SIMPLE      | p     | ALL    | PRIMARY       | NULL    | NULL    | NULL            |      6 | Using where; Using join buffer (Block Nested Loop) |
+----+-------------+-------+--------+---------------+---------+---------+-----------------+--------+----------------------------------------------------+

由于最后 WHERE 条件以及排序均针对最左主表,因此可以先对 my_order 排序提前缩小数据量再做左连接。SQL 重写后如下,执行时间缩小为 1 毫秒左右。

SELECT * 
FROM (
SELECT * 
FROM   my_order o 
WHERE  ( o.display = 0 ) 
       AND ( o.ostaus = 1 ) 
ORDER  BY o.selltime DESC 
LIMIT  0, 15
) o 
     LEFT JOIN my_userinfo u 
              ON o.uid = u.uid 
     LEFT JOIN my_productinfo p 
              ON o.pid = p.pid 
ORDER BY  o.selltime DESC
limit 0, 15

再检查执行计划:子查询物化后(select_type=DERIVED) 参与 JOIN。虽然估算行扫描仍然为 90 万,但是利用了索引以及 LIMIT 子句后,实际执行时间变得很小。

+----+-------------+------------+--------+---------------+---------+---------+-------+--------+----------------------------------------------------+
| id | select_type | table      | type   | possible_keys | key     | key_len | ref   | rows   | Extra                                              |
+----+-------------+------------+--------+---------------+---------+---------+-------+--------+----------------------------------------------------+
|  1 | PRIMARY     | <derived2> | ALL    | NULL          | NULL    | NULL    | NULL  |     15 | Using temporary; Using filesort                    |
|  1 | PRIMARY     | u          | eq_ref | PRIMARY       | PRIMARY | 4       | o.uid |      1 | NULL                                               |
|  1 | PRIMARY     | p          | ALL    | PRIMARY       | NULL    | NULL    | NULL  |      6 | Using where; Using join buffer (Block Nested Loop) |
|  2 | DERIVED     | o          | index  | NULL          | idx_1   | 5       | NULL  | 909112 | Using where                                        |
+----+-------------+------------+--------+---------------+---------+---------+-------+--------+----------------------------------------------------+

8、中间结果集下推

再来看下面这个已经初步优化过的例子 (左连接中的主表优先作用查询条件):

SELECT    a.*, 
          c.allocated 
FROM      ( 
              SELECT   resourceid 
              FROM     my_distribute d 
                   WHERE    isdelete = 0 
                   AND      cusmanagercode = '1234567' 
                   ORDER BY salecode limit 20) a 
LEFT JOIN 
          ( 
              SELECT   resourcesid, sum(ifnull(allocation, 0) * 12345) allocated 
              FROM     my_resources 
                   GROUP BY resourcesid) c 
ON        a.resourceid = c.resourcesid

那么该语句还存在其它问题吗?不难看出子查询 c 是全表聚合查询,在表数量特别大的情况下会导致整个语句的性能下降。

其实对于子查询 c,左连接最后结果集只关心能和主表 resourceid 能匹配的数据。因此我们可以重写语句如下,执行时间从原来的 2 秒下降到 2 毫秒。

SELECT    a.*, 
          c.allocated 
FROM      ( 
                   SELECT   resourceid 
                   FROM     my_distribute d 
                   WHERE    isdelete = 0 
                   AND      cusmanagercode = '1234567' 
                   ORDER BY salecode limit 20) a 
LEFT JOIN 
          ( 
                   SELECT   resourcesid, sum(ifnull(allocation, 0) * 12345) allocated 
                   FROM     my_resources r, 
                            ( 
                                     SELECT   resourceid 
                                     FROM     my_distribute d 
                                     WHERE    isdelete = 0 
                                     AND      cusmanagercode = '1234567' 
                                     ORDER BY salecode limit 20) a 
                   WHERE    r.resourcesid = a.resourcesid 
                   GROUP BY resourcesid) c 
ON        a.resourceid = c.resourcesid

但是子查询 a 在我们的 SQL 语句中出现了多次。这种写法不仅存在额外的开销,还使得整个语句显的繁杂。使用 WITH 语句再次重写:

WITH a AS 
( 
         SELECT   resourceid 
         FROM     my_distribute d 
         WHERE    isdelete = 0 
         AND      cusmanagercode = '1234567' 
         ORDER BY salecode limit 20)
SELECT    a.*, 
          c.allocated 
FROM      a 
LEFT JOIN 
          ( 
                   SELECT   resourcesid, sum(ifnull(allocation, 0) * 12345) allocated 
                   FROM     my_resources r, 
                            a 
                   WHERE    r.resourcesid = a.resourcesid 
                   GROUP BY resourcesid) c 
ON        a.resourceid = c.resourcesid

数据库编译器产生执行计划,决定着 SQL 的实际执行方式。但是编译器只是尽力服务,所有数据库的编译器都不是尽善尽美的。

上述提到的多数场景,在其它数据库中也存在性能问题。了解数据库编译器的特性,才能避规其短处,写出高性能的 SQL 语句。

程序员在设计数据模型以及编写 SQL 语句时,要把算法的思想或意识带进来。

编写复杂 SQL 语句要养成使用 WITH 语句的习惯。简洁且思路清晰的 SQL 语句也能减小数据库的负担 。


“不积跬步,无以至千里”,希望未来的你能:有梦为马 随处可栖!加油,少年!

关注公众号:「Java 知己」,每天更新Java知识哦,期待你的到来!

  • 发送「Group」,与 10 万程序员一起进步。
  • 发送「面试」,领取BATJ面试资料、面试视频攻略。
  • 发送「玩转算法」,领取《玩转算法》系列视频教程。
  • 千万不要发送「1024」...
    每日福利

原文链接:https://www.cnblogs.com/java-friend/p/11881539.html
如有疑问请与原作者联系

标签:

版权申明:本站文章部分自网络,如有侵权,请联系:west999com@outlook.com
特别注意:本站所有转载文章言论不代表本站观点,本站所提供的摄影照片,插画,设计作品,如需使用,请与原作者联系,版权归原作者所有

上一篇:还看不懂同事的代码?超强的 Stream 流操作姿势还不学习一下

下一篇:《吊打面试官》系列-秒杀系统设计